Robot learning of agricultural tasks

With this research we aim to increase the adaptability of agricultural robotics, and thereby increase the automation in agriculture. One of the main challenges for increasing the automatization in agriculture is the large variety of work done on farms. This research will focus on using learning-based control of robots to allow agricultural robotics to perform more tasks. We will focus on the task of tomato harvesting to identify the possibilities and challenges of applying learning-based robotics in agriculture. Due to the complex environment, learning from interaction with the environment is not feasible. We will use learning based on observed expert data, Learning from Demonstration, as it allows the human to demonstrate the specific motions used for detaching tomatoes and thus reduces the time and data required for learning. This research aims to incorporate Learning from Demonstration (LfD) in agricultural robotics. Four studies are proposed, all focusing on tomato harvesting. We focus on the part of tomato harvesting where the hand is already near the tomato. When the hand is close to the tomato, it is then closed around the tomato and the tomato is manipulated to make the detachment from the plant happen. The first two studies focus on the manipulation of the tomato to make the detachment happen, where the pose of the end-effector will be controlled. The last two studies focus on controlling an anthropomorphic soft robotic hand to grasp and hold the tomato.